3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
10 * linux/net/ipv4/tcp_input.c
11 * linux/net/ipv4/tcp_output.c
14 * Hideaki YOSHIFUJI : sin6_scope_id support
15 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
16 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
17 * a single port at the same time.
18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
26 #include <linux/bottom_half.h>
27 #include <linux/module.h>
28 #include <linux/errno.h>
29 #include <linux/types.h>
30 #include <linux/socket.h>
31 #include <linux/sockios.h>
32 #include <linux/net.h>
33 #include <linux/jiffies.h>
35 #include <linux/in6.h>
36 #include <linux/netdevice.h>
37 #include <linux/init.h>
38 #include <linux/jhash.h>
39 #include <linux/ipsec.h>
40 #include <linux/times.h>
41 #include <linux/slab.h>
42 #include <linux/uaccess.h>
43 #include <linux/ipv6.h>
44 #include <linux/icmpv6.h>
45 #include <linux/random.h>
48 #include <net/ndisc.h>
49 #include <net/inet6_hashtables.h>
50 #include <net/inet6_connection_sock.h>
52 #include <net/transp_v6.h>
53 #include <net/addrconf.h>
54 #include <net/ip6_route.h>
55 #include <net/ip6_checksum.h>
56 #include <net/inet_ecn.h>
57 #include <net/protocol.h>
60 #include <net/dsfield.h>
61 #include <net/timewait_sock.h>
62 #include <net/inet_common.h>
63 #include <net/secure_seq.h>
64 #include <net/busy_poll.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
69 #include <crypto/hash.h>
70 #include <linux/scatterlist.h>
72 #include <trace/events/tcp.h>
74 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
75 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
76 struct request_sock *req);
78 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
80 static const struct inet_connection_sock_af_ops ipv6_mapped;
81 static const struct inet_connection_sock_af_ops ipv6_specific;
82 #ifdef CONFIG_TCP_MD5SIG
83 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
84 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
86 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
87 const struct in6_addr *addr)
93 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
95 struct dst_entry *dst = skb_dst(skb);
97 if (dst && dst_hold_safe(dst)) {
98 const struct rt6_info *rt = (const struct rt6_info *)dst;
101 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
102 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
106 static u32 tcp_v6_init_seq(const struct sk_buff *skb)
108 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
109 ipv6_hdr(skb)->saddr.s6_addr32,
111 tcp_hdr(skb)->source);
114 static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
116 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
117 ipv6_hdr(skb)->saddr.s6_addr32);
120 static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
123 /* This check is replicated from tcp_v6_connect() and intended to
124 * prevent BPF program called below from accessing bytes that are out
125 * of the bound specified by user in addr_len.
127 if (addr_len < SIN6_LEN_RFC2133)
130 sock_owned_by_me(sk);
132 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr);
135 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
138 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
139 struct inet_sock *inet = inet_sk(sk);
140 struct inet_connection_sock *icsk = inet_csk(sk);
141 struct ipv6_pinfo *np = inet6_sk(sk);
142 struct tcp_sock *tp = tcp_sk(sk);
143 struct in6_addr *saddr = NULL, *final_p, final;
144 struct ipv6_txoptions *opt;
146 struct dst_entry *dst;
149 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
151 if (addr_len < SIN6_LEN_RFC2133)
154 if (usin->sin6_family != AF_INET6)
155 return -EAFNOSUPPORT;
157 memset(&fl6, 0, sizeof(fl6));
160 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
161 IP6_ECN_flow_init(fl6.flowlabel);
162 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
163 struct ip6_flowlabel *flowlabel;
164 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
167 fl6_sock_release(flowlabel);
172 * connect() to INADDR_ANY means loopback (BSD'ism).
175 if (ipv6_addr_any(&usin->sin6_addr)) {
176 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
177 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
180 usin->sin6_addr = in6addr_loopback;
183 addr_type = ipv6_addr_type(&usin->sin6_addr);
185 if (addr_type & IPV6_ADDR_MULTICAST)
188 if (addr_type&IPV6_ADDR_LINKLOCAL) {
189 if (addr_len >= sizeof(struct sockaddr_in6) &&
190 usin->sin6_scope_id) {
191 /* If interface is set while binding, indices
194 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
197 sk->sk_bound_dev_if = usin->sin6_scope_id;
200 /* Connect to link-local address requires an interface */
201 if (!sk->sk_bound_dev_if)
205 if (tp->rx_opt.ts_recent_stamp &&
206 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
207 tp->rx_opt.ts_recent = 0;
208 tp->rx_opt.ts_recent_stamp = 0;
212 sk->sk_v6_daddr = usin->sin6_addr;
213 np->flow_label = fl6.flowlabel;
219 if (addr_type & IPV6_ADDR_MAPPED) {
220 u32 exthdrlen = icsk->icsk_ext_hdr_len;
221 struct sockaddr_in sin;
223 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
225 if (__ipv6_only_sock(sk))
228 sin.sin_family = AF_INET;
229 sin.sin_port = usin->sin6_port;
230 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
232 icsk->icsk_af_ops = &ipv6_mapped;
233 sk->sk_backlog_rcv = tcp_v4_do_rcv;
234 #ifdef CONFIG_TCP_MD5SIG
235 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
238 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
241 icsk->icsk_ext_hdr_len = exthdrlen;
242 icsk->icsk_af_ops = &ipv6_specific;
243 sk->sk_backlog_rcv = tcp_v6_do_rcv;
244 #ifdef CONFIG_TCP_MD5SIG
245 tp->af_specific = &tcp_sock_ipv6_specific;
249 np->saddr = sk->sk_v6_rcv_saddr;
254 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
255 saddr = &sk->sk_v6_rcv_saddr;
257 fl6.flowi6_proto = IPPROTO_TCP;
258 fl6.daddr = sk->sk_v6_daddr;
259 fl6.saddr = saddr ? *saddr : np->saddr;
260 fl6.flowi6_oif = sk->sk_bound_dev_if;
261 fl6.flowi6_mark = sk->sk_mark;
262 fl6.fl6_dport = usin->sin6_port;
263 fl6.fl6_sport = inet->inet_sport;
264 fl6.flowi6_uid = sk->sk_uid;
266 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
267 final_p = fl6_update_dst(&fl6, opt, &final);
269 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
271 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
279 sk->sk_v6_rcv_saddr = *saddr;
282 /* set the source address */
284 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
286 sk->sk_gso_type = SKB_GSO_TCPV6;
287 ip6_dst_store(sk, dst, NULL, NULL);
289 icsk->icsk_ext_hdr_len = 0;
291 icsk->icsk_ext_hdr_len = opt->opt_flen +
294 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
296 inet->inet_dport = usin->sin6_port;
298 tcp_set_state(sk, TCP_SYN_SENT);
299 err = inet6_hash_connect(tcp_death_row, sk);
305 if (likely(!tp->repair)) {
307 tp->write_seq = secure_tcpv6_seq(np->saddr.s6_addr32,
308 sk->sk_v6_daddr.s6_addr32,
311 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
313 sk->sk_v6_daddr.s6_addr32);
316 if (tcp_fastopen_defer_connect(sk, &err))
321 err = tcp_connect(sk);
328 tcp_set_state(sk, TCP_CLOSE);
330 inet->inet_dport = 0;
331 sk->sk_route_caps = 0;
335 static void tcp_v6_mtu_reduced(struct sock *sk)
337 struct dst_entry *dst;
339 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
342 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
346 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
347 tcp_sync_mss(sk, dst_mtu(dst));
348 tcp_simple_retransmit(sk);
352 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
353 u8 type, u8 code, int offset, __be32 info)
355 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
356 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
357 struct net *net = dev_net(skb->dev);
358 struct request_sock *fastopen;
359 struct ipv6_pinfo *np;
366 sk = __inet6_lookup_established(net, &tcp_hashinfo,
367 &hdr->daddr, th->dest,
368 &hdr->saddr, ntohs(th->source),
369 skb->dev->ifindex, inet6_sdif(skb));
372 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
377 if (sk->sk_state == TCP_TIME_WAIT) {
378 inet_twsk_put(inet_twsk(sk));
381 seq = ntohl(th->seq);
382 fatal = icmpv6_err_convert(type, code, &err);
383 if (sk->sk_state == TCP_NEW_SYN_RECV)
384 return tcp_req_err(sk, seq, fatal);
387 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
388 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
390 if (sk->sk_state == TCP_CLOSE)
393 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
394 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
399 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
400 fastopen = tp->fastopen_rsk;
401 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
402 if (sk->sk_state != TCP_LISTEN &&
403 !between(seq, snd_una, tp->snd_nxt)) {
404 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
410 if (type == NDISC_REDIRECT) {
411 if (!sock_owned_by_user(sk)) {
412 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
415 dst->ops->redirect(dst, sk, skb);
420 if (type == ICMPV6_PKT_TOOBIG) {
421 /* We are not interested in TCP_LISTEN and open_requests
422 * (SYN-ACKs send out by Linux are always <576bytes so
423 * they should go through unfragmented).
425 if (sk->sk_state == TCP_LISTEN)
428 if (!ip6_sk_accept_pmtu(sk))
431 tp->mtu_info = ntohl(info);
432 if (!sock_owned_by_user(sk))
433 tcp_v6_mtu_reduced(sk);
434 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
441 /* Might be for an request_sock */
442 switch (sk->sk_state) {
445 /* Only in fast or simultaneous open. If a fast open socket is
446 * is already accepted it is treated as a connected one below.
448 if (fastopen && !fastopen->sk)
451 if (!sock_owned_by_user(sk)) {
453 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
457 sk->sk_err_soft = err;
461 if (!sock_owned_by_user(sk) && np->recverr) {
463 sk->sk_error_report(sk);
465 sk->sk_err_soft = err;
473 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
475 struct request_sock *req,
476 struct tcp_fastopen_cookie *foc,
477 enum tcp_synack_type synack_type)
479 struct inet_request_sock *ireq = inet_rsk(req);
480 struct ipv6_pinfo *np = inet6_sk(sk);
481 struct ipv6_txoptions *opt;
482 struct flowi6 *fl6 = &fl->u.ip6;
486 /* First, grab a route. */
487 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
488 IPPROTO_TCP)) == NULL)
491 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
494 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
495 &ireq->ir_v6_rmt_addr);
497 fl6->daddr = ireq->ir_v6_rmt_addr;
498 if (np->repflow && ireq->pktopts)
499 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
502 opt = ireq->ipv6_opt;
504 opt = rcu_dereference(np->opt);
505 err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass);
507 err = net_xmit_eval(err);
515 static void tcp_v6_reqsk_destructor(struct request_sock *req)
517 kfree(inet_rsk(req)->ipv6_opt);
518 kfree_skb(inet_rsk(req)->pktopts);
521 #ifdef CONFIG_TCP_MD5SIG
522 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
523 const struct in6_addr *addr)
525 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
528 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
529 const struct sock *addr_sk)
531 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
534 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
535 char __user *optval, int optlen)
537 struct tcp_md5sig cmd;
538 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
541 if (optlen < sizeof(cmd))
544 if (copy_from_user(&cmd, optval, sizeof(cmd)))
547 if (sin6->sin6_family != AF_INET6)
550 if (optname == TCP_MD5SIG_EXT &&
551 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
552 prefixlen = cmd.tcpm_prefixlen;
553 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
557 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
560 if (!cmd.tcpm_keylen) {
561 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
562 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
564 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
565 AF_INET6, prefixlen);
568 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
571 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
572 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
573 AF_INET, prefixlen, cmd.tcpm_key,
574 cmd.tcpm_keylen, GFP_KERNEL);
576 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
577 AF_INET6, prefixlen, cmd.tcpm_key,
578 cmd.tcpm_keylen, GFP_KERNEL);
581 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
582 const struct in6_addr *daddr,
583 const struct in6_addr *saddr,
584 const struct tcphdr *th, int nbytes)
586 struct tcp6_pseudohdr *bp;
587 struct scatterlist sg;
591 /* 1. TCP pseudo-header (RFC2460) */
594 bp->protocol = cpu_to_be32(IPPROTO_TCP);
595 bp->len = cpu_to_be32(nbytes);
597 _th = (struct tcphdr *)(bp + 1);
598 memcpy(_th, th, sizeof(*th));
601 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
602 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
603 sizeof(*bp) + sizeof(*th));
604 return crypto_ahash_update(hp->md5_req);
607 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
608 const struct in6_addr *daddr, struct in6_addr *saddr,
609 const struct tcphdr *th)
611 struct tcp_md5sig_pool *hp;
612 struct ahash_request *req;
614 hp = tcp_get_md5sig_pool();
616 goto clear_hash_noput;
619 if (crypto_ahash_init(req))
621 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
623 if (tcp_md5_hash_key(hp, key))
625 ahash_request_set_crypt(req, NULL, md5_hash, 0);
626 if (crypto_ahash_final(req))
629 tcp_put_md5sig_pool();
633 tcp_put_md5sig_pool();
635 memset(md5_hash, 0, 16);
639 static int tcp_v6_md5_hash_skb(char *md5_hash,
640 const struct tcp_md5sig_key *key,
641 const struct sock *sk,
642 const struct sk_buff *skb)
644 const struct in6_addr *saddr, *daddr;
645 struct tcp_md5sig_pool *hp;
646 struct ahash_request *req;
647 const struct tcphdr *th = tcp_hdr(skb);
649 if (sk) { /* valid for establish/request sockets */
650 saddr = &sk->sk_v6_rcv_saddr;
651 daddr = &sk->sk_v6_daddr;
653 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
654 saddr = &ip6h->saddr;
655 daddr = &ip6h->daddr;
658 hp = tcp_get_md5sig_pool();
660 goto clear_hash_noput;
663 if (crypto_ahash_init(req))
666 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
668 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
670 if (tcp_md5_hash_key(hp, key))
672 ahash_request_set_crypt(req, NULL, md5_hash, 0);
673 if (crypto_ahash_final(req))
676 tcp_put_md5sig_pool();
680 tcp_put_md5sig_pool();
682 memset(md5_hash, 0, 16);
688 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
689 const struct sk_buff *skb)
691 #ifdef CONFIG_TCP_MD5SIG
692 const __u8 *hash_location = NULL;
693 struct tcp_md5sig_key *hash_expected;
694 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
695 const struct tcphdr *th = tcp_hdr(skb);
699 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
700 hash_location = tcp_parse_md5sig_option(th);
702 /* We've parsed the options - do we have a hash? */
703 if (!hash_expected && !hash_location)
706 if (hash_expected && !hash_location) {
707 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
711 if (!hash_expected && hash_location) {
712 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
716 /* check the signature */
717 genhash = tcp_v6_md5_hash_skb(newhash,
721 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
722 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
723 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
724 genhash ? "failed" : "mismatch",
725 &ip6h->saddr, ntohs(th->source),
726 &ip6h->daddr, ntohs(th->dest));
733 static void tcp_v6_init_req(struct request_sock *req,
734 const struct sock *sk_listener,
737 struct inet_request_sock *ireq = inet_rsk(req);
738 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
740 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
741 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
743 /* So that link locals have meaning */
744 if (!sk_listener->sk_bound_dev_if &&
745 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
746 ireq->ir_iif = tcp_v6_iif(skb);
748 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
749 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
750 np->rxopt.bits.rxinfo ||
751 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
752 np->rxopt.bits.rxohlim || np->repflow)) {
753 refcount_inc(&skb->users);
758 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
760 const struct request_sock *req)
762 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
765 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
767 .obj_size = sizeof(struct tcp6_request_sock),
768 .rtx_syn_ack = tcp_rtx_synack,
769 .send_ack = tcp_v6_reqsk_send_ack,
770 .destructor = tcp_v6_reqsk_destructor,
771 .send_reset = tcp_v6_send_reset,
772 .syn_ack_timeout = tcp_syn_ack_timeout,
775 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
776 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
777 sizeof(struct ipv6hdr),
778 #ifdef CONFIG_TCP_MD5SIG
779 .req_md5_lookup = tcp_v6_md5_lookup,
780 .calc_md5_hash = tcp_v6_md5_hash_skb,
782 .init_req = tcp_v6_init_req,
783 #ifdef CONFIG_SYN_COOKIES
784 .cookie_init_seq = cookie_v6_init_sequence,
786 .route_req = tcp_v6_route_req,
787 .init_seq = tcp_v6_init_seq,
788 .init_ts_off = tcp_v6_init_ts_off,
789 .send_synack = tcp_v6_send_synack,
792 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
793 u32 ack, u32 win, u32 tsval, u32 tsecr,
794 int oif, struct tcp_md5sig_key *key, int rst,
795 u8 tclass, __be32 label)
797 const struct tcphdr *th = tcp_hdr(skb);
799 struct sk_buff *buff;
801 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
802 struct sock *ctl_sk = net->ipv6.tcp_sk;
803 unsigned int tot_len = sizeof(struct tcphdr);
804 struct dst_entry *dst;
809 tot_len += TCPOLEN_TSTAMP_ALIGNED;
810 #ifdef CONFIG_TCP_MD5SIG
812 tot_len += TCPOLEN_MD5SIG_ALIGNED;
815 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
820 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
822 t1 = skb_push(buff, tot_len);
823 skb_reset_transport_header(buff);
825 /* Swap the send and the receive. */
826 memset(t1, 0, sizeof(*t1));
827 t1->dest = th->source;
828 t1->source = th->dest;
829 t1->doff = tot_len / 4;
830 t1->seq = htonl(seq);
831 t1->ack_seq = htonl(ack);
832 t1->ack = !rst || !th->ack;
834 t1->window = htons(win);
836 topt = (__be32 *)(t1 + 1);
839 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
840 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
841 *topt++ = htonl(tsval);
842 *topt++ = htonl(tsecr);
845 #ifdef CONFIG_TCP_MD5SIG
847 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
848 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
849 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
850 &ipv6_hdr(skb)->saddr,
851 &ipv6_hdr(skb)->daddr, t1);
855 memset(&fl6, 0, sizeof(fl6));
856 fl6.daddr = ipv6_hdr(skb)->saddr;
857 fl6.saddr = ipv6_hdr(skb)->daddr;
858 fl6.flowlabel = label;
860 buff->ip_summed = CHECKSUM_PARTIAL;
863 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
865 fl6.flowi6_proto = IPPROTO_TCP;
866 if (rt6_need_strict(&fl6.daddr) && !oif)
867 fl6.flowi6_oif = tcp_v6_iif(skb);
869 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
872 fl6.flowi6_oif = oif;
876 mark = (sk->sk_state == TCP_TIME_WAIT) ?
877 inet_twsk(sk)->tw_mark : sk->sk_mark;
878 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
879 fl6.fl6_dport = t1->dest;
880 fl6.fl6_sport = t1->source;
881 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
882 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
884 /* Pass a socket to ip6_dst_lookup either it is for RST
885 * Underlying function will use this to retrieve the network
888 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
890 skb_dst_set(buff, dst);
891 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, tclass);
892 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
894 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
901 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
903 const struct tcphdr *th = tcp_hdr(skb);
904 u32 seq = 0, ack_seq = 0;
905 struct tcp_md5sig_key *key = NULL;
906 #ifdef CONFIG_TCP_MD5SIG
907 const __u8 *hash_location = NULL;
908 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
909 unsigned char newhash[16];
911 struct sock *sk1 = NULL;
918 /* If sk not NULL, it means we did a successful lookup and incoming
919 * route had to be correct. prequeue might have dropped our dst.
921 if (!sk && !ipv6_unicast_destination(skb))
924 #ifdef CONFIG_TCP_MD5SIG
926 hash_location = tcp_parse_md5sig_option(th);
927 if (sk && sk_fullsock(sk)) {
928 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
929 } else if (hash_location) {
931 * active side is lost. Try to find listening socket through
932 * source port, and then find md5 key through listening socket.
933 * we are not loose security here:
934 * Incoming packet is checked with md5 hash with finding key,
935 * no RST generated if md5 hash doesn't match.
937 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
938 &tcp_hashinfo, NULL, 0,
940 th->source, &ipv6h->daddr,
941 ntohs(th->source), tcp_v6_iif(skb),
946 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
950 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
951 if (genhash || memcmp(hash_location, newhash, 16) != 0)
957 seq = ntohl(th->ack_seq);
959 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
963 oif = sk->sk_bound_dev_if;
965 trace_tcp_send_reset(sk, skb);
968 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
970 #ifdef CONFIG_TCP_MD5SIG
976 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
977 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
978 struct tcp_md5sig_key *key, u8 tclass,
981 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
985 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
987 struct inet_timewait_sock *tw = inet_twsk(sk);
988 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
990 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
991 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
992 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
993 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
994 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
999 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1000 struct request_sock *req)
1002 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1003 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1006 * The window field (SEG.WND) of every outgoing segment, with the
1007 * exception of <SYN> segments, MUST be right-shifted by
1008 * Rcv.Wind.Shift bits:
1010 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1011 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1012 tcp_rsk(req)->rcv_nxt,
1013 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1014 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1015 req->ts_recent, sk->sk_bound_dev_if,
1016 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr),
1021 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1023 #ifdef CONFIG_SYN_COOKIES
1024 const struct tcphdr *th = tcp_hdr(skb);
1027 sk = cookie_v6_check(sk, skb);
1032 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1034 if (skb->protocol == htons(ETH_P_IP))
1035 return tcp_v4_conn_request(sk, skb);
1037 if (!ipv6_unicast_destination(skb))
1040 return tcp_conn_request(&tcp6_request_sock_ops,
1041 &tcp_request_sock_ipv6_ops, sk, skb);
1045 return 0; /* don't send reset */
1048 static void tcp_v6_restore_cb(struct sk_buff *skb)
1050 /* We need to move header back to the beginning if xfrm6_policy_check()
1051 * and tcp_v6_fill_cb() are going to be called again.
1052 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1054 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1055 sizeof(struct inet6_skb_parm));
1058 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1059 struct request_sock *req,
1060 struct dst_entry *dst,
1061 struct request_sock *req_unhash,
1064 struct inet_request_sock *ireq;
1065 struct ipv6_pinfo *newnp;
1066 const struct ipv6_pinfo *np = inet6_sk(sk);
1067 struct ipv6_txoptions *opt;
1068 struct tcp6_sock *newtcp6sk;
1069 struct inet_sock *newinet;
1070 struct tcp_sock *newtp;
1072 #ifdef CONFIG_TCP_MD5SIG
1073 struct tcp_md5sig_key *key;
1077 if (skb->protocol == htons(ETH_P_IP)) {
1082 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1083 req_unhash, own_req);
1088 newtcp6sk = (struct tcp6_sock *)newsk;
1089 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1091 newinet = inet_sk(newsk);
1092 newnp = inet6_sk(newsk);
1093 newtp = tcp_sk(newsk);
1095 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1097 newnp->saddr = newsk->sk_v6_rcv_saddr;
1099 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1100 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1101 #ifdef CONFIG_TCP_MD5SIG
1102 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1105 newnp->ipv6_mc_list = NULL;
1106 newnp->ipv6_ac_list = NULL;
1107 newnp->ipv6_fl_list = NULL;
1108 newnp->pktoptions = NULL;
1110 newnp->mcast_oif = tcp_v6_iif(skb);
1111 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1112 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1114 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1117 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1118 * here, tcp_create_openreq_child now does this for us, see the comment in
1119 * that function for the gory details. -acme
1122 /* It is tricky place. Until this moment IPv4 tcp
1123 worked with IPv6 icsk.icsk_af_ops.
1126 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1131 ireq = inet_rsk(req);
1133 if (sk_acceptq_is_full(sk))
1137 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1142 newsk = tcp_create_openreq_child(sk, req, skb);
1147 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1148 * count here, tcp_create_openreq_child now does this for us, see the
1149 * comment in that function for the gory details. -acme
1152 newsk->sk_gso_type = SKB_GSO_TCPV6;
1153 ip6_dst_store(newsk, dst, NULL, NULL);
1154 inet6_sk_rx_dst_set(newsk, skb);
1156 newtcp6sk = (struct tcp6_sock *)newsk;
1157 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1159 newtp = tcp_sk(newsk);
1160 newinet = inet_sk(newsk);
1161 newnp = inet6_sk(newsk);
1163 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1165 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1166 newnp->saddr = ireq->ir_v6_loc_addr;
1167 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1168 newsk->sk_bound_dev_if = ireq->ir_iif;
1170 /* Now IPv6 options...
1172 First: no IPv4 options.
1174 newinet->inet_opt = NULL;
1175 newnp->ipv6_mc_list = NULL;
1176 newnp->ipv6_ac_list = NULL;
1177 newnp->ipv6_fl_list = NULL;
1180 newnp->rxopt.all = np->rxopt.all;
1182 newnp->pktoptions = NULL;
1184 newnp->mcast_oif = tcp_v6_iif(skb);
1185 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1186 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1188 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1190 /* Clone native IPv6 options from listening socket (if any)
1192 Yes, keeping reference count would be much more clever,
1193 but we make one more one thing there: reattach optmem
1196 opt = ireq->ipv6_opt;
1198 opt = rcu_dereference(np->opt);
1200 opt = ipv6_dup_options(newsk, opt);
1201 RCU_INIT_POINTER(newnp->opt, opt);
1203 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1205 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1208 tcp_ca_openreq_child(newsk, dst);
1210 tcp_sync_mss(newsk, dst_mtu(dst));
1211 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1213 tcp_initialize_rcv_mss(newsk);
1215 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1216 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1218 #ifdef CONFIG_TCP_MD5SIG
1219 /* Copy over the MD5 key from the original socket */
1220 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1222 /* We're using one, so create a matching key
1223 * on the newsk structure. If we fail to get
1224 * memory, then we end up not copying the key
1227 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1228 AF_INET6, 128, key->key, key->keylen,
1229 sk_gfp_mask(sk, GFP_ATOMIC));
1233 if (__inet_inherit_port(sk, newsk) < 0) {
1234 inet_csk_prepare_forced_close(newsk);
1238 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1240 tcp_move_syn(newtp, req);
1242 /* Clone pktoptions received with SYN, if we own the req */
1243 if (ireq->pktopts) {
1244 newnp->pktoptions = skb_clone(ireq->pktopts,
1245 sk_gfp_mask(sk, GFP_ATOMIC));
1246 consume_skb(ireq->pktopts);
1247 ireq->pktopts = NULL;
1248 if (newnp->pktoptions) {
1249 tcp_v6_restore_cb(newnp->pktoptions);
1250 skb_set_owner_r(newnp->pktoptions, newsk);
1258 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1266 /* The socket must have it's spinlock held when we get
1267 * here, unless it is a TCP_LISTEN socket.
1269 * We have a potential double-lock case here, so even when
1270 * doing backlog processing we use the BH locking scheme.
1271 * This is because we cannot sleep with the original spinlock
1274 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1276 struct ipv6_pinfo *np = inet6_sk(sk);
1277 struct tcp_sock *tp;
1278 struct sk_buff *opt_skb = NULL;
1280 /* Imagine: socket is IPv6. IPv4 packet arrives,
1281 goes to IPv4 receive handler and backlogged.
1282 From backlog it always goes here. Kerboom...
1283 Fortunately, tcp_rcv_established and rcv_established
1284 handle them correctly, but it is not case with
1285 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1288 if (skb->protocol == htons(ETH_P_IP))
1289 return tcp_v4_do_rcv(sk, skb);
1292 * socket locking is here for SMP purposes as backlog rcv
1293 * is currently called with bh processing disabled.
1296 /* Do Stevens' IPV6_PKTOPTIONS.
1298 Yes, guys, it is the only place in our code, where we
1299 may make it not affecting IPv4.
1300 The rest of code is protocol independent,
1301 and I do not like idea to uglify IPv4.
1303 Actually, all the idea behind IPV6_PKTOPTIONS
1304 looks not very well thought. For now we latch
1305 options, received in the last packet, enqueued
1306 by tcp. Feel free to propose better solution.
1310 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1312 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1313 struct dst_entry *dst = sk->sk_rx_dst;
1315 sock_rps_save_rxhash(sk, skb);
1316 sk_mark_napi_id(sk, skb);
1318 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1319 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1321 sk->sk_rx_dst = NULL;
1325 tcp_rcv_established(sk, skb);
1327 goto ipv6_pktoptions;
1331 if (tcp_checksum_complete(skb))
1334 if (sk->sk_state == TCP_LISTEN) {
1335 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1341 if (tcp_child_process(sk, nsk, skb))
1344 __kfree_skb(opt_skb);
1348 sock_rps_save_rxhash(sk, skb);
1350 if (tcp_rcv_state_process(sk, skb))
1353 goto ipv6_pktoptions;
1357 tcp_v6_send_reset(sk, skb);
1360 __kfree_skb(opt_skb);
1364 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1365 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1370 /* Do you ask, what is it?
1372 1. skb was enqueued by tcp.
1373 2. skb is added to tail of read queue, rather than out of order.
1374 3. socket is not in passive state.
1375 4. Finally, it really contains options, which user wants to receive.
1378 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1379 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1380 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1381 np->mcast_oif = tcp_v6_iif(opt_skb);
1382 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1383 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1384 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1385 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1387 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1388 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1389 skb_set_owner_r(opt_skb, sk);
1390 tcp_v6_restore_cb(opt_skb);
1391 opt_skb = xchg(&np->pktoptions, opt_skb);
1393 __kfree_skb(opt_skb);
1394 opt_skb = xchg(&np->pktoptions, NULL);
1402 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1403 const struct tcphdr *th)
1405 /* This is tricky: we move IP6CB at its correct location into
1406 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1407 * _decode_session6() uses IP6CB().
1408 * barrier() makes sure compiler won't play aliasing games.
1410 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1411 sizeof(struct inet6_skb_parm));
1414 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1415 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1416 skb->len - th->doff*4);
1417 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1418 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1419 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1420 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1421 TCP_SKB_CB(skb)->sacked = 0;
1422 TCP_SKB_CB(skb)->has_rxtstamp =
1423 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1426 static int tcp_v6_rcv(struct sk_buff *skb)
1428 int sdif = inet6_sdif(skb);
1429 const struct tcphdr *th;
1430 const struct ipv6hdr *hdr;
1434 struct net *net = dev_net(skb->dev);
1436 if (skb->pkt_type != PACKET_HOST)
1440 * Count it even if it's bad.
1442 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1444 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1447 th = (const struct tcphdr *)skb->data;
1449 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1451 if (!pskb_may_pull(skb, th->doff*4))
1454 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1457 th = (const struct tcphdr *)skb->data;
1458 hdr = ipv6_hdr(skb);
1461 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1462 th->source, th->dest, inet6_iif(skb), sdif,
1468 if (sk->sk_state == TCP_TIME_WAIT)
1471 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1472 struct request_sock *req = inet_reqsk(sk);
1473 bool req_stolen = false;
1476 sk = req->rsk_listener;
1477 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1478 sk_drops_add(sk, skb);
1482 if (tcp_checksum_complete(skb)) {
1486 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1487 inet_csk_reqsk_queue_drop_and_put(sk, req);
1493 if (!tcp_filter(sk, skb)) {
1494 th = (const struct tcphdr *)skb->data;
1495 hdr = ipv6_hdr(skb);
1496 tcp_v6_fill_cb(skb, hdr, th);
1497 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1502 /* Another cpu got exclusive access to req
1503 * and created a full blown socket.
1504 * Try to feed this packet to this socket
1505 * instead of discarding it.
1507 tcp_v6_restore_cb(skb);
1511 goto discard_and_relse;
1515 tcp_v6_restore_cb(skb);
1516 } else if (tcp_child_process(sk, nsk, skb)) {
1517 tcp_v6_send_reset(nsk, skb);
1518 goto discard_and_relse;
1524 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1525 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1526 goto discard_and_relse;
1529 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1530 goto discard_and_relse;
1532 if (tcp_v6_inbound_md5_hash(sk, skb))
1533 goto discard_and_relse;
1535 if (tcp_filter(sk, skb))
1536 goto discard_and_relse;
1537 th = (const struct tcphdr *)skb->data;
1538 hdr = ipv6_hdr(skb);
1539 tcp_v6_fill_cb(skb, hdr, th);
1543 if (sk->sk_state == TCP_LISTEN) {
1544 ret = tcp_v6_do_rcv(sk, skb);
1545 goto put_and_return;
1548 sk_incoming_cpu_update(sk);
1550 bh_lock_sock_nested(sk);
1551 tcp_segs_in(tcp_sk(sk), skb);
1553 if (!sock_owned_by_user(sk)) {
1554 ret = tcp_v6_do_rcv(sk, skb);
1555 } else if (tcp_add_backlog(sk, skb)) {
1556 goto discard_and_relse;
1563 return ret ? -1 : 0;
1566 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1569 tcp_v6_fill_cb(skb, hdr, th);
1571 if (tcp_checksum_complete(skb)) {
1573 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1575 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1577 tcp_v6_send_reset(NULL, skb);
1585 sk_drops_add(sk, skb);
1591 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1592 inet_twsk_put(inet_twsk(sk));
1596 tcp_v6_fill_cb(skb, hdr, th);
1598 if (tcp_checksum_complete(skb)) {
1599 inet_twsk_put(inet_twsk(sk));
1603 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1608 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1609 skb, __tcp_hdrlen(th),
1610 &ipv6_hdr(skb)->saddr, th->source,
1611 &ipv6_hdr(skb)->daddr,
1612 ntohs(th->dest), tcp_v6_iif(skb),
1615 struct inet_timewait_sock *tw = inet_twsk(sk);
1616 inet_twsk_deschedule_put(tw);
1618 tcp_v6_restore_cb(skb);
1626 tcp_v6_timewait_ack(sk, skb);
1629 tcp_v6_send_reset(sk, skb);
1630 inet_twsk_deschedule_put(inet_twsk(sk));
1632 case TCP_TW_SUCCESS:
1638 static void tcp_v6_early_demux(struct sk_buff *skb)
1640 const struct ipv6hdr *hdr;
1641 const struct tcphdr *th;
1644 if (skb->pkt_type != PACKET_HOST)
1647 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1650 hdr = ipv6_hdr(skb);
1653 if (th->doff < sizeof(struct tcphdr) / 4)
1656 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1657 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1658 &hdr->saddr, th->source,
1659 &hdr->daddr, ntohs(th->dest),
1660 inet6_iif(skb), inet6_sdif(skb));
1663 skb->destructor = sock_edemux;
1664 if (sk_fullsock(sk)) {
1665 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1668 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1670 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1671 skb_dst_set_noref(skb, dst);
1676 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1677 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1678 .twsk_unique = tcp_twsk_unique,
1679 .twsk_destructor = tcp_twsk_destructor,
1682 static const struct inet_connection_sock_af_ops ipv6_specific = {
1683 .queue_xmit = inet6_csk_xmit,
1684 .send_check = tcp_v6_send_check,
1685 .rebuild_header = inet6_sk_rebuild_header,
1686 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1687 .conn_request = tcp_v6_conn_request,
1688 .syn_recv_sock = tcp_v6_syn_recv_sock,
1689 .net_header_len = sizeof(struct ipv6hdr),
1690 .net_frag_header_len = sizeof(struct frag_hdr),
1691 .setsockopt = ipv6_setsockopt,
1692 .getsockopt = ipv6_getsockopt,
1693 .addr2sockaddr = inet6_csk_addr2sockaddr,
1694 .sockaddr_len = sizeof(struct sockaddr_in6),
1695 #ifdef CONFIG_COMPAT
1696 .compat_setsockopt = compat_ipv6_setsockopt,
1697 .compat_getsockopt = compat_ipv6_getsockopt,
1699 .mtu_reduced = tcp_v6_mtu_reduced,
1702 #ifdef CONFIG_TCP_MD5SIG
1703 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1704 .md5_lookup = tcp_v6_md5_lookup,
1705 .calc_md5_hash = tcp_v6_md5_hash_skb,
1706 .md5_parse = tcp_v6_parse_md5_keys,
1711 * TCP over IPv4 via INET6 API
1713 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1714 .queue_xmit = ip_queue_xmit,
1715 .send_check = tcp_v4_send_check,
1716 .rebuild_header = inet_sk_rebuild_header,
1717 .sk_rx_dst_set = inet_sk_rx_dst_set,
1718 .conn_request = tcp_v6_conn_request,
1719 .syn_recv_sock = tcp_v6_syn_recv_sock,
1720 .net_header_len = sizeof(struct iphdr),
1721 .setsockopt = ipv6_setsockopt,
1722 .getsockopt = ipv6_getsockopt,
1723 .addr2sockaddr = inet6_csk_addr2sockaddr,
1724 .sockaddr_len = sizeof(struct sockaddr_in6),
1725 #ifdef CONFIG_COMPAT
1726 .compat_setsockopt = compat_ipv6_setsockopt,
1727 .compat_getsockopt = compat_ipv6_getsockopt,
1729 .mtu_reduced = tcp_v4_mtu_reduced,
1732 #ifdef CONFIG_TCP_MD5SIG
1733 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1734 .md5_lookup = tcp_v4_md5_lookup,
1735 .calc_md5_hash = tcp_v4_md5_hash_skb,
1736 .md5_parse = tcp_v6_parse_md5_keys,
1740 /* NOTE: A lot of things set to zero explicitly by call to
1741 * sk_alloc() so need not be done here.
1743 static int tcp_v6_init_sock(struct sock *sk)
1745 struct inet_connection_sock *icsk = inet_csk(sk);
1749 icsk->icsk_af_ops = &ipv6_specific;
1751 #ifdef CONFIG_TCP_MD5SIG
1752 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1758 static void tcp_v6_destroy_sock(struct sock *sk)
1760 tcp_v4_destroy_sock(sk);
1761 inet6_destroy_sock(sk);
1764 #ifdef CONFIG_PROC_FS
1765 /* Proc filesystem TCPv6 sock list dumping. */
1766 static void get_openreq6(struct seq_file *seq,
1767 const struct request_sock *req, int i)
1769 long ttd = req->rsk_timer.expires - jiffies;
1770 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1771 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1777 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1778 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1780 src->s6_addr32[0], src->s6_addr32[1],
1781 src->s6_addr32[2], src->s6_addr32[3],
1782 inet_rsk(req)->ir_num,
1783 dest->s6_addr32[0], dest->s6_addr32[1],
1784 dest->s6_addr32[2], dest->s6_addr32[3],
1785 ntohs(inet_rsk(req)->ir_rmt_port),
1787 0, 0, /* could print option size, but that is af dependent. */
1788 1, /* timers active (only the expire timer) */
1789 jiffies_to_clock_t(ttd),
1791 from_kuid_munged(seq_user_ns(seq),
1792 sock_i_uid(req->rsk_listener)),
1793 0, /* non standard timer */
1794 0, /* open_requests have no inode */
1798 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1800 const struct in6_addr *dest, *src;
1803 unsigned long timer_expires;
1804 const struct inet_sock *inet = inet_sk(sp);
1805 const struct tcp_sock *tp = tcp_sk(sp);
1806 const struct inet_connection_sock *icsk = inet_csk(sp);
1807 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1811 dest = &sp->sk_v6_daddr;
1812 src = &sp->sk_v6_rcv_saddr;
1813 destp = ntohs(inet->inet_dport);
1814 srcp = ntohs(inet->inet_sport);
1816 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1817 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
1818 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1820 timer_expires = icsk->icsk_timeout;
1821 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1823 timer_expires = icsk->icsk_timeout;
1824 } else if (timer_pending(&sp->sk_timer)) {
1826 timer_expires = sp->sk_timer.expires;
1829 timer_expires = jiffies;
1832 state = inet_sk_state_load(sp);
1833 if (state == TCP_LISTEN)
1834 rx_queue = sp->sk_ack_backlog;
1836 /* Because we don't lock the socket,
1837 * we might find a transient negative value.
1839 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1842 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1843 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1845 src->s6_addr32[0], src->s6_addr32[1],
1846 src->s6_addr32[2], src->s6_addr32[3], srcp,
1847 dest->s6_addr32[0], dest->s6_addr32[1],
1848 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1850 tp->write_seq - tp->snd_una,
1853 jiffies_delta_to_clock_t(timer_expires - jiffies),
1854 icsk->icsk_retransmits,
1855 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1856 icsk->icsk_probes_out,
1858 refcount_read(&sp->sk_refcnt), sp,
1859 jiffies_to_clock_t(icsk->icsk_rto),
1860 jiffies_to_clock_t(icsk->icsk_ack.ato),
1861 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1863 state == TCP_LISTEN ?
1864 fastopenq->max_qlen :
1865 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1869 static void get_timewait6_sock(struct seq_file *seq,
1870 struct inet_timewait_sock *tw, int i)
1872 long delta = tw->tw_timer.expires - jiffies;
1873 const struct in6_addr *dest, *src;
1876 dest = &tw->tw_v6_daddr;
1877 src = &tw->tw_v6_rcv_saddr;
1878 destp = ntohs(tw->tw_dport);
1879 srcp = ntohs(tw->tw_sport);
1882 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1883 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1885 src->s6_addr32[0], src->s6_addr32[1],
1886 src->s6_addr32[2], src->s6_addr32[3], srcp,
1887 dest->s6_addr32[0], dest->s6_addr32[1],
1888 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1889 tw->tw_substate, 0, 0,
1890 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1891 refcount_read(&tw->tw_refcnt), tw);
1894 static int tcp6_seq_show(struct seq_file *seq, void *v)
1896 struct tcp_iter_state *st;
1897 struct sock *sk = v;
1899 if (v == SEQ_START_TOKEN) {
1904 "st tx_queue rx_queue tr tm->when retrnsmt"
1905 " uid timeout inode\n");
1910 if (sk->sk_state == TCP_TIME_WAIT)
1911 get_timewait6_sock(seq, v, st->num);
1912 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1913 get_openreq6(seq, v, st->num);
1915 get_tcp6_sock(seq, v, st->num);
1920 static const struct seq_operations tcp6_seq_ops = {
1921 .show = tcp6_seq_show,
1922 .start = tcp_seq_start,
1923 .next = tcp_seq_next,
1924 .stop = tcp_seq_stop,
1927 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1931 int __net_init tcp6_proc_init(struct net *net)
1933 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
1934 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
1939 void tcp6_proc_exit(struct net *net)
1941 remove_proc_entry("tcp6", net->proc_net);
1945 struct proto tcpv6_prot = {
1947 .owner = THIS_MODULE,
1949 .pre_connect = tcp_v6_pre_connect,
1950 .connect = tcp_v6_connect,
1951 .disconnect = tcp_disconnect,
1952 .accept = inet_csk_accept,
1954 .init = tcp_v6_init_sock,
1955 .destroy = tcp_v6_destroy_sock,
1956 .shutdown = tcp_shutdown,
1957 .setsockopt = tcp_setsockopt,
1958 .getsockopt = tcp_getsockopt,
1959 .keepalive = tcp_set_keepalive,
1960 .recvmsg = tcp_recvmsg,
1961 .sendmsg = tcp_sendmsg,
1962 .sendpage = tcp_sendpage,
1963 .backlog_rcv = tcp_v6_do_rcv,
1964 .release_cb = tcp_release_cb,
1966 .unhash = inet_unhash,
1967 .get_port = inet_csk_get_port,
1968 .enter_memory_pressure = tcp_enter_memory_pressure,
1969 .leave_memory_pressure = tcp_leave_memory_pressure,
1970 .stream_memory_free = tcp_stream_memory_free,
1971 .sockets_allocated = &tcp_sockets_allocated,
1972 .memory_allocated = &tcp_memory_allocated,
1973 .memory_pressure = &tcp_memory_pressure,
1974 .orphan_count = &tcp_orphan_count,
1975 .sysctl_mem = sysctl_tcp_mem,
1976 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
1977 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
1978 .max_header = MAX_TCP_HEADER,
1979 .obj_size = sizeof(struct tcp6_sock),
1980 .slab_flags = SLAB_TYPESAFE_BY_RCU,
1981 .twsk_prot = &tcp6_timewait_sock_ops,
1982 .rsk_prot = &tcp6_request_sock_ops,
1983 .h.hashinfo = &tcp_hashinfo,
1984 .no_autobind = true,
1985 #ifdef CONFIG_COMPAT
1986 .compat_setsockopt = compat_tcp_setsockopt,
1987 .compat_getsockopt = compat_tcp_getsockopt,
1989 .diag_destroy = tcp_abort,
1992 /* thinking of making this const? Don't.
1993 * early_demux can change based on sysctl.
1995 static struct inet6_protocol tcpv6_protocol = {
1996 .early_demux = tcp_v6_early_demux,
1997 .early_demux_handler = tcp_v6_early_demux,
1998 .handler = tcp_v6_rcv,
1999 .err_handler = tcp_v6_err,
2000 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2003 static struct inet_protosw tcpv6_protosw = {
2004 .type = SOCK_STREAM,
2005 .protocol = IPPROTO_TCP,
2006 .prot = &tcpv6_prot,
2007 .ops = &inet6_stream_ops,
2008 .flags = INET_PROTOSW_PERMANENT |
2012 static int __net_init tcpv6_net_init(struct net *net)
2014 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2015 SOCK_RAW, IPPROTO_TCP, net);
2018 static void __net_exit tcpv6_net_exit(struct net *net)
2020 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2023 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2025 inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2028 static struct pernet_operations tcpv6_net_ops = {
2029 .init = tcpv6_net_init,
2030 .exit = tcpv6_net_exit,
2031 .exit_batch = tcpv6_net_exit_batch,
2034 int __init tcpv6_init(void)
2038 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2042 /* register inet6 protocol */
2043 ret = inet6_register_protosw(&tcpv6_protosw);
2045 goto out_tcpv6_protocol;
2047 ret = register_pernet_subsys(&tcpv6_net_ops);
2049 goto out_tcpv6_protosw;
2054 inet6_unregister_protosw(&tcpv6_protosw);
2056 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2060 void tcpv6_exit(void)
2062 unregister_pernet_subsys(&tcpv6_net_ops);
2063 inet6_unregister_protosw(&tcpv6_protosw);
2064 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);